DSets-DBSCAN: A Parameter-Free Clustering Algorithm.

Area of Science:

• Computer Vision
• Machine Learning
• Data Science

Background:

• Image segmentation and data clustering are vital techniques.
• Existing algorithms often require user-tuned parameters, limiting practical application.
• Parameter dependence can hinder reproducible and efficient clustering results.

Purpose of the Study:

• To develop a parameter-free clustering algorithm for image segmentation and data clustering.
• To overcome the limitations of user-specified parameters in clustering methods.
• To enhance the robustness and applicability of clustering techniques.

Main Methods:

• A novel parameter-free algorithm integrating dominant sets (DSets) and Density-Based Spatial Clustering of Applications with Noise (DBSCAN).
• Application of histogram equalization to the pairwise similarity matrix for parameter independence in DSets.
• Automatic determination of DBSCAN parameters based on DSets clustering outputs.
• Extension of DSets clusters using DBSCAN to form arbitrary-shaped clusters.

Main Results:

• The proposed algorithm successfully generates clusters of arbitrary shapes without requiring user input parameters.
• Experimental results in data clustering and image segmentation demonstrate superior or comparable performance against existing methods.
• The algorithm effectively removes the dependence on user-specified parameters, enhancing practical usability.

Conclusions:

• The parameter-free algorithm effectively merges the strengths of DSets and DBSCAN for robust clustering.
• This approach offers a significant advancement in automated image segmentation and data clustering.
• The method provides a reliable and efficient alternative to parameter-dependent clustering algorithms.